Bagging

Clase series de tiempo, primavera 2024

library(tidyquant)
library(tidyverse)
library(tsibble)
library(fable)
library(feasts)

Data

data = tq_get(c("AMZN", "DIS"), get = "stock.prices", from = "2023-01-01", to = "2024-03-29") %>% select(symbol, date, adjusted)
data

data = data %>% group_by(symbol) %>% mutate(t = row_number())
data

data = as_tsibble(data, index = t, key = symbol)

Model spec

fit = data %>% model(stl = STL(adjusted))
fit |>
  components() |>
  autoplot()

filter(fit, symbol == "DIS") %>% gg_tsresiduals()

filter(fit, symbol == "DIS") %>%
  generate(new_data = data, times = 10,
           bootstrap_block_size = 8) |>
  autoplot(.sim) +
  autolayer(filter(data, symbol=="DIS"), adjusted, colour = "black") +
  guides(colour = "none") +
  labs(title = "Stock prices: Bootstrapped series",
       y="price")

Warning: no non-missing arguments to max; returning -Inf

filter(fit, symbol == "AMZN") %>%
  generate(new_data = data, times = 10,
           bootstrap_block_size = 100) |>
  autoplot(.sim) +
  autolayer(filter(data, symbol=="AMZN"), adjusted, colour = "black") +
  guides(colour = "none") +
  labs(title = "Stock prices: Bootstrapped series",
       y="price")

Warning: no non-missing arguments to max; returning -Inf

Simulación

sim <- fit |>
  generate(new_data = data, times = 100,
           bootstrap_block_size = 50) %>%
  select(-.model, -adjusted)

Warning: no non-missing arguments to max; returning -InfWarning: no non-missing arguments to max; returning -Inf

sim

arima_forecasts <- sim |>
  model(arima = ARIMA(.sim)) |>
  forecast(h = 20)

Warning: NaNs producedWarning: NaNs producedWarning: NaNs producedWarning: NaNs producedWarning: NaNs produced

arima_forecasts |> filter(symbol=="DIS") %>%
  as_tsibble(key = c(".rep", "symbol"), index = t) %>%
  autoplot(.mean) +
  autolayer(filter(data, symbol=="DIS"), adjusted, colour = "black") +
  guides(colour = "none") +
  labs(title = "Stock prices: Bootstrapped series",y="price", x = "period")

bagged <- ets_forecasts |>  group_by(symbol) %>%
  summarise(bagged_mean = mean(.mean))

data |> filter(symbol=="AMZN") %>%
  autoplot() +
  autolayer(filter(bagged, symbol=="AMZN")) +
  labs(title = "Stock prices: Bootstrapped series",y="price", x = "period")

Plot variable not specified, automatically selected `.vars = adjusted`Plot variable not specified, automatically selected `.vars = bagged_mean`

Ensamble

STLF <- decomposition_model(
  STL(adjusted ~ season(window = Inf)),
  ETS(season_adjust ~ season("N"))
)

fit_ensamble = data %>% model(
                            ets = ETS(adjusted),
                            arima = ARIMA(adjusted),
                            STLF
) %>% mutate(combination = (ets + arima + STLF)/3)

fc = fit_ensamble %>% forecast(h = 20)

fit_ensamble

accuracy(fit_ensamble) %>% arrange(MAPE)

fc %>% autoplot(data, level=NULL)

`mutate_if()` ignored the following grouping variables:

ticker = "DIS"
fit_ensamble %>% filter(symbol == ticker) %>% augment() |>
  ggplot(aes(x = t)) +
  geom_line(aes(y = adjusted, colour = "reales")) +
  geom_line(aes(y = .fitted, colour = .model)) +
  labs(y = NULL,
    title = ticker
  ) +
  guides(colour = guide_legend(title = NULL))